Table 1.
Microstructural classification results using object-based CNN and pixel-based MVFCNN approaches.
| Method | Type | Training Strategy | Accuracy |
|---|---|---|---|
| Pauly et al.21 | object-based | — | 48.89% |
| CIFAR-Net | object-based | from scratch | 57.03% |
| pre-trained VGG19-Net Features + SVM | object-based | — | 64.84% |
| VGG16-Net | object-based | fine tuning | 66.50% |
| MVFCNN | pixel-based | fine tuning | 93.94% |
The results show that object-based classification approaches improve over prior work at most by around 18 percent points. The pixel-based approach has even better performance by around 45 percent points improvement.